Spatial Projection of Multiple Climate Variables Using Hierarchical Multitask Learning

Authors: Andre Goncalves, Arindam Banerjee, Fernando Von Zuben

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on synthetic and real climate data show that HMTL produces better results than decoupled MTL methods applied separately on the super-tasks and HMTL significantly outperforms baselines for climate projection.
Researcher Affiliation Collaboration Andr e R. Gonc alves Center for Research and Development in Telecommunication (CPq D), Brazil andrerg@cpqd.com.br Arindam Banerjee Dept. of Comp. Sci. & Engg. University of Minnesota, USA banerjee@cs.umn.edu Fernando J. Von Zuben School of Elec. and Comp. Eng. University of Campinas, Brazil vonzuben@dca.fee.unicamp.br
Pseudocode Yes Algorithm 1: HMTL algorithm. Data: {X}, {Y}. Input: λ0 > 0, λ1 > 0 and λ2 > 0. Result: {Θ}, {Ω}. 2 Ω(t) = Imt, t = 1, ..., T. 3 Θ(t) = U( 0.5, 0.5), t = 1, ..., T. 5 Update {Θ} by solving (5); 6 Update {Ω} by solving (6); 7 until stopping condition met
Open Source Code No The paper does not explicitly state that source code for the methodology is available or provide a link to a repository.
Open Datasets Yes We collected monthly land temperature and precipitation data of 32 CMIP5 ESMs (Taylor, Stouffer, and Meehl 2012), from 1901 to 2000, in South America. Observed data provided by (Willmott and Matsuura 2001) was used.
Dataset Splits Yes All the penalization parameters of the methods (λ s in MSSL and HMTL) were chosen by cross-validation. From the training set, we selected the first 80% for training and the next 20% for validation. The best values in the validation set were selected.
Hardware Specification No The paper does not provide any specific details about the hardware used for experiments.
Software Dependencies No The paper mentions optimization methods like L-BFGS and ADMM, but does not specify any software names with version numbers or other software dependencies.
Experiment Setup Yes All the penalization parameters of the methods (λ s in MSSL and HMTL) were chosen by cross-validation. ... Using this protocol, the selected parameter values were: S2M2R used λ = 1000; MSSL λ0 = 0.1 and λ1 = 0.1; and HMTL λ0 = 0.1, λ1 = 0.0002, λ2 = 0.01.